133 lines
3.4 KiB
Odin
133 lines
3.4 KiB
Odin
// General mathematical constructions used for the prototype
|
||
|
||
package sectr
|
||
|
||
import "core:math"
|
||
|
||
// These are the same as the runtime constants for memory units just using a more general name when not refering to bytes
|
||
|
||
Kilo :: Kilobyte
|
||
Mega :: Megabyte
|
||
Giga :: Gigabyte
|
||
Tera :: Terabyte
|
||
Peta :: Petabyte
|
||
Exa :: Exabyte
|
||
|
||
Axis2 :: enum i32 {
|
||
Invalid = -1,
|
||
X = 0,
|
||
Y = 1,
|
||
Count,
|
||
}
|
||
|
||
f32_Infinity :: 0x7F800000
|
||
f32_Min :: 0x00800000
|
||
|
||
// Note(Ed) : I don't see an intrinsict available anywhere for this. So I'll be using the Terathon non-sse impl
|
||
// Inverse Square Root
|
||
// C++ Source https://github.com/EricLengyel/Terathon-Math-Library/blob/main/TSMath.cpp#L191
|
||
inverse_sqrt_f32 :: proc "contextless" ( value : f32 ) -> f32
|
||
{
|
||
if ( value < f32_Min) {
|
||
return f32_Infinity
|
||
}
|
||
|
||
value_u32 := transmute(u32) value
|
||
|
||
initial_approx := 0x5F375A86 - (value_u32 >> 1)
|
||
refined_approx := transmute(f32) initial_approx
|
||
|
||
// Newton–Raphson method for getting better approximations of square roots
|
||
// Done twice for greater accuracy.
|
||
refined_approx = refined_approx * (1.5 - value * 0.5 * refined_approx * refined_approx )
|
||
refined_approx = refined_approx * (1.5 - value * 0.5 * refined_approx * refined_approx )
|
||
// refined_approx = (0.5 * refined_approx) * (3.0 - value * refined_approx * refined_approx)
|
||
// refined_approx = (0.5 * refined_approx) * (3.0 - value * refined_approx * refined_approx)
|
||
return refined_approx
|
||
}
|
||
|
||
is_power_of_two_u32 :: #force_inline proc "contextless" ( value : u32 ) -> b32
|
||
{
|
||
return value != 0 && ( value & ( value - 1 )) == 0
|
||
}
|
||
|
||
mov_avg_exp_f32 := #force_inline proc "contextless" ( alpha, delta_interval, last_value : f32 ) -> f32
|
||
{
|
||
result := (delta_interval * alpha) + (delta_interval * (1.0 - alpha))
|
||
return result
|
||
}
|
||
|
||
mov_avg_exp_f64 := #force_inline proc "contextless" ( alpha, delta_interval, last_value : f64 ) -> f64
|
||
{
|
||
result := (delta_interval * alpha) + (delta_interval * (1.0 - alpha))
|
||
return result
|
||
}
|
||
|
||
import "core:math/linalg"
|
||
|
||
Quat128 :: quaternion128
|
||
Matrix2 :: matrix [2, 2] f32
|
||
Vec2i :: [2]i32
|
||
Vec3i :: [3]i32
|
||
|
||
vec2i_to_vec2 :: #force_inline proc "contextless" (v : Vec2i) -> Vec2 {return transmute(Vec2) v}
|
||
vec3i_to_vec3 :: #force_inline proc "contextless" (v : Vec3i) -> Vec3 {return transmute(Vec3) v}
|
||
|
||
#region("Range2")
|
||
|
||
Range2 :: struct #raw_union {
|
||
using min_max : struct {
|
||
min, max : Vec2
|
||
},
|
||
using pts : struct {
|
||
p0, p1 : Vec2
|
||
},
|
||
using xy : struct {
|
||
x0, y0 : f32,
|
||
x1, y1 : f32,
|
||
},
|
||
using side : struct {
|
||
left, bottom : f32,
|
||
right, top : f32,
|
||
},
|
||
ratio : struct {
|
||
x, y : f32,
|
||
},
|
||
|
||
// TODO(Ed) : Test these
|
||
array : [4]f32,
|
||
mat : matrix[2, 2] f32,
|
||
}
|
||
|
||
UnitRange2 :: distinct Range2
|
||
|
||
range2 :: #force_inline proc "contextless" ( a, b : Vec2 ) -> Range2 {
|
||
result := Range2 { pts = { a, b } }
|
||
return result
|
||
}
|
||
|
||
add_range2 :: #force_inline proc "contextless" ( a, b : Range2 ) -> Range2 {
|
||
result := Range2 { pts = {
|
||
a.p0 + b.p0,
|
||
a.p1 + b.p1,
|
||
}}
|
||
return result
|
||
}
|
||
|
||
sub_range2 :: #force_inline proc "contextless" ( a, b : Range2 ) -> Range2 {
|
||
// result := Range2 { array = a.array - b.array }
|
||
result := Range2 { mat = a.mat - b.mat }
|
||
return result
|
||
}
|
||
|
||
equal_range2 :: #force_inline proc "contextless" ( a, b : Range2 ) -> b32 {
|
||
result := a.p0 == b.p0 && a.p1 == b.p1
|
||
return b32(result)
|
||
}
|
||
|
||
size_range2 :: #force_inline proc "contextless" ( value : Range2 ) -> Vec2 {
|
||
return { value.p1.x - value.p0.x, value.p0.y - value.p1.y }
|
||
}
|
||
|
||
#endregion("Range2")
|